GCU MOSTLY DETERMINISTIC / REGISTRY: FLORIDA / COMMISSIONED 2026 / STATUS: COMMISSIONING
Mostly Deterministic designs and builds agentic AI platforms for enterprises. Architecture, engineering, evaluation, and the guardrails that let a probabilistic model do a dependable job.
> MIND:The same input will produce the same output. Probably.
Six capabilities, one unit. Each is delivered as working software, not a slide.
The foundation an enterprise needs to build and run agents at scale: orchestration, retrieval, grounding, memory, and tool integration.
Agents and the components around them, built hands-on: multi-agent orchestration, RAG pipelines, function calling, and Model Context Protocol integrations.
Choosing and routing between models, and engineering the prompts and context that decide how they behave.
Testing, evaluation, and monitoring that show whether an agent is doing its job, before launch and after.
Identity, access control, auditability, data governance, and responsible AI requirements built in from the first commit.
Working with architecture, security, and platform teams so the system fits the organization it has to live in.
> MIND:Confidence figures are illustrative. That one is 1.00.
An agent that hasn't been evaluated is a rumor. Evaluation is built alongside the feature.
Agents get the access they need and no more. Every action leaves a trail.
Novel models, conventional engineering. The interesting part should be the capability, never the outage.
The client's team should be able to operate, extend, and question everything after the unit departs.
Designs and builds AI agent platforms for large organizations, from architecture through production code. Works at the intersection of software architecture and applied AI, with a focus on agentic infrastructure, LLM systems, and voice.
SISTER SHIPAlso aboard the GSV Well, It Compiled, a separate vessel. See Vertrus.
> MIND:The org chart is a dot.
Software is supposed to be deterministic. Language models are not. The work happens in the gap: engineering around a probabilistic core until the whole thing is dependable enough to run a business on.
Mostly.
A nod to the ship Minds of Iain M. Banks's Culture novels, who choose their own names.
Building an agent platform, or trying to make one behave? Send a signal. Replies come from the one human aboard.